Pro-Code
For developers who want hands-on time building agents.
About the session
Retrieval-Augmented Generation (RAG) has transformed how AI applications access external knowledge, but reliable AI systems require more than retrieving documents and generating responses. Building trustworthy agents means giving models the ability to reason across multiple sources, select the right tools, evaluate evidence, and clearly communicate where their answers come from. In this technical session, Hastimal will demonstrate an open-source healthcare information discovery platform built with Gemini, Google’s Agent Development Kit (ADK), and publicly accessible healthcare data. Using a real-world scenario involving a parent searching for pediatric dental providers in Houston for an anxious child, attendees will see how specialized agents interpret complex requests, decompose them into smaller tasks, retrieve provider information from NPPES, query biomedical literature through PubMed, and synthesize the results into transparent, source-cited responses. The session focuses on practical agentic architecture and information-discovery patterns rather than medical advice, with an emphasis on evidence grounding, multi-agent orchestration, tool use, and responsible AI design.
About Hastimal
Hastimal Jangid is a Cloud & AI Architect and Co-Founder & CEO of RankRabbit AI, where he focuses on AI visibility and reputation management for growing businesses across search, LLMs, local, and social platforms. An IEEE Senior Member and Google Developer Expert (GDE), he brings more than 15 years of experience across technology and education. His technical work spans Google Cloud, Gemini, Google ADK, Agentic AI, and cloud-native architectures, with a focus on building practical, production-ready AI systems. Hastimal is also passionate about open source and developer education, regularly sharing hands-on technical knowledge through conference talks, workshops, community programs, and collaborative projects.
About the session
Production AI agents fail in a specific way: they run unbounded, with no enforced budget, no approval step before a consequential tool call, no audit trail, and no way to replay a run to find where it went wrong. Pollard is an open-source library for Python, Node, and Rust that wraps agent execution in governed, hash-linked execution trees: budgets on tokens, cost, and steps are enforced while the agent runs, high-risk actions pause at human approval gates, and any completed run can be replayed exactly for debugging and audit. In this hands-on workshop, you’ll wrap a working agent in a governed execution tree, set budgets and watch them enforced live, add an approval gate in front of a destructive tool call, replay a recorded run to diagnose a failure, and connect the Pollard MCP server so the same governance applies to MCP-based tooling.
About Muntaser
A Lead Gen AI Engineer at Insight Global and former Technical Lead at NVIDIA, specializing in trustworthy AI, autonomous agents, and machine learning. An award-winning innovator and open-source contributor, he conducts research at Florida Institute of Technology.
About the session
Tools like Claude Code and Cursor feel like magic, but at their core is a simple loop wrapped in careful engineering, called the “harness.” In this hands-on workshop, attendees will build a working coding agent from scratch in Python that can read files, run commands, and fix a failing test. Step by step, we’ll add tool calling, context compaction, skills, and tracing with the open-source observability platform Langfuse. We’ll finish by comparing our hand-built harness to production frameworks like the Claude Agent SDK and LangChain, so attendees understand what these frameworks do and when to use them.
About Ashwini
A data science and machine learning professional at Sumo Logic with experience in AI, natural language processing, and predictive analytics. She holds a graduate degree from Duke University and has worked on projects spanning deep learning, recommendation systems, and cloud-based AI applications.








